app.py

from llama_index.core import (
    VectorStoreIndex
)
from llama_index.core import Settings
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_index.vector_stores.qdrant import QdrantVectorStore
from qdrant_client import QdrantClient
from typing import Any, List, Tuple
import torch
from transformers import AutoTokenizer, AutoModelForMaskedLM
import streamlit as st
from llama_index.llms.huggingface import (
    HuggingFaceInferenceAPI
)
import os
HUGGINGFACEHUB_API_TOKEN = os.environ.get("HUGGINGFACEHUB_API_TOKEN")
Q_END_POINT = os.environ.get("Q_END_POINT")
Q_API_KEY = os.environ.get("Q_API_KEY")


#DOC
#https://docs.llamaindex.ai/en/stable/examples/vector_stores/qdrant_hybrid.html

doc_tokenizer = AutoTokenizer.from_pretrained(
    "naver/efficient-splade-VI-BT-large-doc"
)
doc_model = AutoModelForMaskedLM.from_pretrained(
    "naver/efficient-splade-VI-BT-large-doc"
)

query_tokenizer = AutoTokenizer.from_pretrained(
    "naver/efficient-splade-VI-BT-large-query"
)
query_model = AutoModelForMaskedLM.from_pretrained(
    "naver/efficient-splade-VI-BT-large-query"
)

device = "cuda:0" if torch.cuda.is_available() else "cpu"

doc_model = doc_model.to(device)
query_model = query_model.to(device)


def sparse_doc_vectors(
    texts: List[str],
) -> Tuple[List[List[int]], List[List[float]]]:
    """
    Computes vectors from logits and attention mask using ReLU, log, and max operations.
    """
    tokens = doc_tokenizer(
        texts, truncation=True, padding=True, return_tensors="pt"
    )
    if torch.cuda.is_available():
        tokens = tokens.to("cuda:1")

    output = doc_model(**tokens)
    logits, attention_mask = output.logits, tokens.attention_mask
    relu_log = torch.log(1 + torch.relu(logits))
    weighted_log = relu_log * attention_mask.unsqueeze(-1)
    tvecs, _ = torch.max(weighted_log, dim=1)

    # extract the vectors that are non-zero and their indices
    indices = []
    vecs = []
    for batch in tvecs:
        indices.append(batch.nonzero(as_tuple=True)[0].tolist())
        vecs.append(batch[indices[-1]].tolist())

    return indices, vecs


def sparse_query_vectors(
    texts: List[str],
) -> Tuple[List[List[int]], List[List[float]]]:
    """
    Computes vectors from logits and attention mask using ReLU, log, and max operations.
    """
    # TODO: compute sparse vectors in batches if max length is exceeded
    tokens = query_tokenizer(
        texts, truncation=True, padding=True, return_tensors="pt"
    )
    if torch.cuda.is_available():
        tokens = tokens.to("cuda:1")


    output = query_model(**tokens)
    logits, attention_mask = output.logits, tokens.attention_mask
    relu_log = torch.log(1 + torch.relu(logits))
    weighted_log = relu_log * attention_mask.unsqueeze(-1)
    tvecs, _ = torch.max(weighted_log, dim=1)

    # extract the vectors that are non-zero and their indices
    indices = []
    vecs = []
    for batch in tvecs:
        indices.append(batch.nonzero(as_tuple=True)[0].tolist())
        vecs.append(batch[indices[-1]].tolist())

    return indices, vecs

st.header("Chat with the Bhagavad Gita docs 💬 📚")

if "messages" not in st.session_state.keys(): # Initialize the chat message history
    st.session_state.messages = [
        {"role": "assistant", "content": "Ask me a question about Gita!"}
    ]


# creates a persistant index to disk
client = QdrantClient(
        Q_END_POINT,
        api_key=Q_API_KEY,
    )
# create our vector store with hybrid indexing enabled
# batch_size controls how many nodes are encoded with sparse vectors at once
vector_store = QdrantVectorStore(
    "bhagavad_gita", client=client, enable_hybrid=True, batch_size=20,force_disable_check_same_thread=True,
    sparse_doc_fn=sparse_doc_vectors,
    sparse_query_fn=sparse_query_vectors,
)


llm = HuggingFaceInferenceAPI(
    model_name="meta-llama/Meta-Llama-3-8B-Instruct", 
    token=HUGGINGFACEHUB_API_TOKEN,
    context_window=8096, 
)
Settings.llm = llm
Settings.tokenzier = AutoTokenizer.from_pretrained(
    "meta-llama/Meta-Llama-3-8B-Instruct"
)

embed_model = HuggingFaceEmbedding(model_name="BAAI/bge-base-en-v1.5", device="cpu")
Settings.embed_model = embed_model

index = VectorStoreIndex.from_vector_store(vector_store=vector_store,embed_model=embed_model)

from llama_index.core.memory import ChatMemoryBuffer
memory = ChatMemoryBuffer.from_defaults(token_limit=1500)

chat_engine = index.as_chat_engine(chat_mode="condense_question", 
                                   verbose=True,
                                   memory=memory,            
                                   sparse_top_k=10,
                                   vector_store_query_mode="hybrid",
                                   similarity_top_k=3,
                                   )

if prompt := st.chat_input("Your question"): # Prompt for user input and save to chat history
    st.session_state.messages.append({"role": "user", "content": prompt})

for message in st.session_state.messages: # Display the prior chat messages
    with st.chat_message(message["role"]):
        st.write(message["content"])

# If last message is not from assistant, generate a new response
if st.session_state.messages[-1]["role"] != "assistant":
    with st.chat_message("assistant"):
        with st.spinner("Thinking..."):
            response = chat_engine.chat(prompt)
            st.write(response.response)
            message = {"role": "assistant", "content": response.response}
            st.session_state.messages.append(message) # Add response to message history